Generator-Linked Provenance as Security-Relevant Side Information: A Frozen-History Protocol for Vulnerability Prioritization in AI-Generated Code

Code-generation systems may exhibit recurring implementation patterns associated with security weaknesses. Whether recorded generation provenance improves vulnerability prioritization beyond technical code features is a separate empirical question. This paper specifies a three-part frozen-history comparison: a historical partition constructs pseudonymous generator profiles, a ranker partition supports fitting and validation, and a held-out evaluation partition supports paired comparison. Each generation event produces one designated function in a fixed non-generated harness. The primary target is retrieval of functions with at least one adjudicated security weakness. The primary queue contains outputs for one task across generators and therefore measures cross-generator portfolio selection, not ordinary single-generator review. A registered secondary analysis evaluates within-generator workloads across tasks. Three matched procedures separate technical features, identity/configuration, and frozen history. An R-only attribution probe measures generator recoverability from the technical representation without treating recoverability as proof of security redundancy. Assessment eligibility is separate from security state, and reference negatives are not described as secure. Risk models address provenance corruption, leakage, assessment circularity, workload mismatch, statistical overclaim, and misuse. A companion local calculator evaluates imported fitted scores; software checks do not establish scientific effectiveness. The study estimates paired out-of-sample ranking differences, not causal generator effects, universal signatures, or practical attacker advantage.

Authors

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-16
DOI
https://doi.org/10.5281/zenodo.24268566
Primary Topic
Software Engineering Research
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Generator-Linked Provenance as Security-Relevant Side Information: A Frozen-History Protocol for Vulnerability Prioritization in AI-Generated Code

Narnaiezzsshaa Truong
1 links
Zenodo (CERN European Organization for Nuclear Research)
Software Engineering Research
article

Generator-Linked Provenance as Security-Relevant Side Information: A Frozen-History Protocol for Vulnerability Prioritization in AI-Generated Code

Narnaiezzsshaa Truong
article en

Abstract

Code-generation systems may exhibit recurring implementation patterns associated with security weaknesses. Whether recorded generation provenance improves vulnerability prioritization beyond technical code features is a separate empirical question. This paper specifies a three-part frozen-history comparison: a historical partition constructs pseudonymous generator profiles, a ranker partition supports fitting and validation, and a held-out evaluation partition supports paired comparison. Each generation event produces one designated function in a fixed non-generated harness. The primary target is retrieval of functions with at least one adjudicated security weakness. The primary queue contains outputs for one task across generators and therefore measures cross-generator portfolio selection, not ordinary single-generator review. A registered secondary analysis evaluates within-generator workloads across tasks. Three matched procedures separate technical features, identity/configuration, and frozen history. An R-only attribution probe measures generator recoverability from the technical representation without treating recoverability as proof of security redundancy. Assessment eligibility is separate from security state, and reference negatives are not described as secure. Risk models address provenance corruption, leakage, assessment circularity, workload mismatch, statistical overclaim, and misuse. A companion local calculator evaluates imported fitted scores; software checks do not establish scientific effectiveness. The study estimates paired out-of-sample ranking differences, not causal generator effects, universal signatures, or practical attacker advantage.

Zenodo (CERN European Organization for Nuclear Research)
Software Engineering Research
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon